Upload skypilot.yaml with huggingface_hub

skypilot.yaml

@@ -0,0 +1,72 @@
+# SkyPilot YAML — PARLE Light Pipeline
+# Qwen 0.5B + gTTS (minimal VRAM)
+#
+# Usage:
+#   sky launch -c parle-light skypilot.yaml --cloud vast --gpus RTX3090:1 -i 5 --down -y
+#   sky launch -c parle-light skypilot.yaml --cloud runpod --gpus A4000:1 -i 5 --down -y
+#
+# Stack:
+#   STT: Faster Whisper Small (CUDA float16) — ~460MB VRAM
+#   LLM: Qwen2.5-0.5B-Instruct (float16)    — ~1GB VRAM
+#   TTS: gTTS (Google API, zero GPU)
+#
+# Total VRAM: ~1.5 GB (runs on any GPU)
+# Cold start: ~3–5 min
+
+resources:
+  cloud: vast
+  accelerators: RTX3090:1
+  disk_size: 50
+  memory: 16+
+  use_spot: true
+  ports:
+    - 8000  # FastAPI server (HTTP/SSE + WebSocket)
+
+envs:
+  HF_TOKEN: null
+
+setup: |
+  set -e
+
+  echo "=== Cloning PARLE Light from HuggingFace ==="
+  if [ -n "$HF_TOKEN" ]; then
+    git clone https://hf-user:${HF_TOKEN}@huggingface.co/spaces/marcosremar2/parle-s2s-light ~/parle-speech
+  else
+    git clone https://huggingface.co/spaces/marcosremar2/parle-s2s-light ~/parle-speech
+  fi
+
+  echo "=== Installing system dependencies ==="
+  sudo apt-get update -qq
+  sudo apt-get install -y -qq ffmpeg libsndfile1 python3-pip git
+
+  echo "=== Installing Python dependencies ==="
+  cd ~/parle-speech
+
+  pip install --upgrade pip setuptools wheel
+
+  # PyTorch with CUDA
+  pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
+
+  # Install all requirements
+  pip install -r requirements.txt
+
+  # Download models (~2 GB total)
+  echo "=== Downloading models ==="
+  python3 ~/parle-speech/download_models.py
+
+  echo "=== Setup complete ==="
+
+run: |
+  set -e
+
+  cd ~/parle-speech
+
+  echo "=== Starting PARLE Light Server ==="
+  echo "Endpoint: http://$(curl -s ifconfig.me):8000"
+  echo "Health: GET http://$(curl -s ifconfig.me):8000/health"
+
+  uvicorn app:app \
+    --host 0.0.0.0 \
+    --port 8000 \
+    --workers 1 \
+    --log-level info